|Publication number||US7913022 B1|
|Application number||US 11/706,556|
|Publication date||Mar 22, 2011|
|Priority date||Feb 14, 2007|
|Publication number||11706556, 706556, US 7913022 B1, US 7913022B1, US-B1-7913022, US7913022 B1, US7913022B1|
|Inventors||Glenn A. Baxter|
|Original Assignee||Xilinx, Inc.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (39), Non-Patent Citations (15), Referenced by (80), Classifications (10), Legal Events (2)|
|External Links: USPTO, USPTO Assignment, Espacenet|
Embodiments of the present invention as described herein are related to the following patent applications, all of which are incorporated by reference herein in their entirety:
U.S. patent application Ser. No. 10/824,967, entitled “Method And Apparatus For Controlling Access To Memory Circuitry,” with inventors Glenn A. Baxter et al., filed Apr. 15, 2004;
U.S. patent application Ser. No. 10/824,713, entitled “Method And Apparatus For Controlling Direct Access To Memory Circuitry,” with inventors Glenn A. Baxter et al., filed Apr. 15, 2004;
U.S. patent application Ser. No. 10/824,715, entitled “Method And Apparatus For Communicating Data Between A Network Transceiver And Memory,” with inventors Christopher J. Borrelli et al., filed Apr. 15, 2004; and
U.S. patent application Ser. No. 11/341,003, entitled “Architecture For Dynamically Programmable Arbitration Using Memory,” with inventor Jennifer R. Lilley, filed Jan. 27, 2006.
1. Technical Field
Embodiments of the present invention relate to features of a versatile Multi-Port Memory Controller (MPMC) that can be included in a system to control access to memory from processors, devices, or buses.
2. Related Art
A conventional processor-based system includes a processor along with a memory, and one or more peripheral devices. The memory stores data and instructions for the computing system. The peripheral devices can include components such as graphics cards, keyboard interfaces, and network interface cards. The computing system can include a system bus to facilitate communication among the processor and peripheral devices and the memory.
With memory access provided through a system bus providing for processor and peripheral devices, arbitration must be performed to gain access to ports of the bus. However, on a shared bus, arbitration is a serial process. That is, a component must request bus access, be granted bus access to the exclusion of all other components, and then perform a memory transaction. The bus arbitration overhead may not allow the full bandwidth capabilities of the memory to be utilized. For instance, the memory is not being kept busy during the time when components are requesting and receiving access to the system bus.
Conventional processor-based systems use some form of memory controller in order to access memory devices and provide arbitration to the memory for the processor and peripherals. Requirements for a memory controller to communicate with different type components and bus structures can decrease bandwidth from the normal operation of a bus-based system. To address the need to configure a memory controller to provide maximum bandwidth when used with various processor systems, a programmable logic device such as an Field Programmable Gate Array (FPGA) has been used to create the memory controller. FPGAs can be used to provide a wide variety of these memory controllers, including single port and multi port memory controllers.
The IOBs 106, the CLBs 104, and the programmable interconnects 108 may be configured to perform a variety of functions. Notably, the CLBs 104 are programmably connectable to each other, and to the IOBs 106, via the programmable interconnect 108. Each CLB slice in turn includes various circuits, such as flip-flops, function generators (e.g., look-up tables (LUTs)), logic gates, and memory. The IOBs 106 are configured to provide input to, and receive output from, the CLBs 104.
Configuration information for the CLBs 104, the IOBs 106, and the programmable interconnect 108 is stored in the configuration memory 116. The configuration memory 116 can include static random access memory (SRAM) cells. A configuration bit stream to program the configuration memory 116 can be produced from the program memory 120.
The IOBs 106 can include transceiver circuitry configured for communication over any of a variety of media, such as wired, wireless, and photonic, whether analog or digital. The DCM blocks 112 provide well-known clock management circuits for managing clock signals within the FPGA 102, such as delay lock loop (DLL) circuits and multiply/divide/de-skew clock circuits.
The processor block 114 comprises a microprocessor core, and typically associated control logic. Notably, such a microprocessor core may include embedded hardware or embedded firmware or a combination. A soft microprocessor 134 may be implemented using the programmable logic of the FPGA 102 (e.g., CLBs 104 and IOBs 106).
As one example, the FPGA used to make an MPMC can be one selected from the Virtex-4 family of products, commercially available from Xilinx, Inc. of San Jose, Calif.
To enable high data-rate communications (e.g., 1200 megabits per second full duplex), the FPGA can be configured as an MPMC with built-in arbitration logic. A typical MPMC will have a fixed number of ports to communicate with components connecting to a memory device. For example, the MPMC may include a port for communicating directly with a central processing unit (CPU) (e.g., an instruction-side processor local bus) and/or a port for communicating with a system bus.
Current MPMC designs have performance issues because of their fixed or non-flexible implementation or architecture. Notably, the systems have a fixed implementation because the port types cannot be changed, and the number of ports remains fixed. Further, they have a fixed arbifration scheme. The systems tend to have port connections to two buses, one for high-speed entities, and one for low-speed entities. The implementation of each of these entities affects the performance such that the lowest performing device on each bus sets the highest frequency possible on that bus. Some system ports are typically dedicated for connection to a CDMAC to allow for direct memory access. Current systems therefore can suffer performance degradation depending on design constraints.
It is desirable to define topologies to efficiently use the components of an FPGA to develop a memory controller. In particular, it is desirable to provide an MPMC that can allow source code to be efficiently changed dynamically to handle a desired number of ports, while maximizing system performance and providing compatibility with a number of different components including peripherals and memory devices that can be connected to the memory controller.
According to embodiments of the present invention, a universal memory controller is provided that can be dynamically made compatible with multiple types of memory as well as multiple types of memory system organizations. Different system topologies are provided through configurable logic. The MPMC configuration provided offers substantially higher bandwidth to devices because of its architecture. It offers freedom of implementation so that customers can trade off area and performance. One implementation uses the resources provided by FPGA technology to efficiently implement the MPMC based system topology.
In some embodiments of the present invention, an MPMC is provided with a self-aligning programmable state machine to provide dynamic compatibility between the MPMC and various memory devices. The self-alignment is provided using shift register look up tables (SRLs) connected to the output of the BRAM state machine forming a part of the MPMC controller. Unlike the fixed implementation of a BRAM-only state machine of conventional systems, this state machine is capable of being updated both dynamically, and statically by providing a delay input to the SRLs and/or updating its contents. This alignment is a key to supporting the dynamic nature of adaptation to various memory organizations, speeds, and timing requirements.
In other embodiments of the present invention, configurable Port Interface Modules (PIMs) are provided on MPMCs ports to enable programmable connection to different type devices, processors, or buses. The PIMs include logic that can be programmed to provide functions ranging from the simple function of a simple direct link, such as a native port interface (NPI), to a more complex DMA such as a CDMAC. The PIMs allow for communication within a wealth of different system topologies. The PIM can further include protocol bridges to allow the PIM to communicate with other PIMs to form a master, slave or master/slave port or a combination thereof. As a master port, the PIM can shift communications from one port to the PIM of another port that provides the necessary translation from a device connected. In this manner, efficient use of the ports is made, and ports will be more available for memory access. Further, buses typically used to communicate with specialized dedicated ports will not be needed with the programmable PIMs, increasing overall port operation speed and reducing latency. Even with bridges external to the MPMC, the PIMs allow a processor to communicate with peripheral devices on different buses while not slowing down remaining devices on a bus communicating with memory over the MPMC. The processor can simultaneously perform instruction reads from memory while communicating with other peripheral devices.
In further embodiments, the data path, address path and PHY interfaces of the MPMC are made highly configurable. In the data path, FIFOs according to embodiments of the present invention are alternatively formed from BRAMs, SRLs, LUT RAM or registers to accommodate different device parameters and operation speed. For example when a PLB typically requires a larger FIFO, a BRAM can be used, while a simple register can be used for an OPB. Further, operation speed at the front end can be adjusted to a different speed than the back end of the FIFO. In the address path, the addresses according to embodiments of the invention can be selectively multiplexed depending on the type memory used. For example, multiplexing can be provided with DRAM memory, and eliminated with SRAM where it is not usually required. Further, the address size can be dynamically adjusted. If an overlap between addresses of two ports occurs, in one embodiment addresses can be semaphored for protection or separation to prevent collisions. In another embodiment, aliasing of addresses can be provided so that when address size is adjusted, the physical addresses do not overlap. For the PHY interface, rather than a fixed device with registers and clocking, the PHY interface according to the present invention is parameterized to align with different memory device types. Instead of using a state machine to read a data address strobe (DAS), embodiments of the invention use logic to read and write a header, payload and footer of data words, compare the values, and adjust input data delays to provide precise data alignment relative to the clock.
In still further embodiments of the present invention, intelligent pipelining is provided to allow control over a variety of pipeline stages in order to permit a customer to trade off FPGA area for maximum frequency performance. The pipelines can be optionally added for each port internal to the MPMC. In one example, the optional pipeline is used between the arbiter and control state machine of the MPMC for each port. The locations of these optional pipelines and their ability to be controlled on a per port basis, achieves greater flexibility and performance.
The MPMC implemented using an FPGA can be programmed in some embodiments through a Graphical User Interface (GUI). The various embodiments of the GUI offer advantageous control over creation of prior MPMC-based systems. First, the GUI allows creation of multiple cores of a system rather than a single core, with an MPMC set to connect the core devices to a particular memory. The GUI can be data driven from user editable text files, and it can offer feedback on resource utilization, area, performance estimates, and performance measurements. It can provide performance feedback estimates based upon the current programmable settings for each port. The GUI can further provide performance data measured for the system back to the user. The GUI can intelligently set the arbitration of the system in such a way as to maximize system performance based upon the measured data. Further, the GUI can use the information entered to dynamically create an entire core for the MPMC and peripherals and processors on board the FPGA, and provide intelligent design rule checking both as information is being entered, as well as during operation when the GUI dynamically creates complex hardware.
In still other embodiments, performance monitors (PMs) can be embedded within the MPMC to provide measurement of various aspects of performance. The measurements can be aggregated together to provide historical information. Additionally, the information collected can be used to dynamically alter the arbitration or any other relevant adjustable parameter within the MPMC. In one embodiment, the PM provides the performance measurement to an external agent for later additional processing and/or summarization.
Further details of the present invention are explained with the help of the attached drawings in which:
The ports 222 are pre-configured with I/O paths capable of communicating with the various types of buses or point-to-point interfaces. Notably, each of the I/O paths of ports 222 0-222 4 include a data path interface (Data) 215, a control bus interface (Cntl) 219, and address bus interface (Addr) 218. The MPMC further includes port arbitration logic 216, data path logic 218, address path logic 212, and control logic 214. The data path logic 218 includes an interface to the data path interface 215 and to the memory 206. The address path logic 212 includes an input interface coupled to the address bus 218 and a memory interface coupled to external memory 150. The port arbitration logic 216 includes an interface coupled to the control bus 219, an interface coupled to the control logic 214, an interface coupled to the data path logic 218, and an interface coupled to the address path logic 212. The control logic 214 includes a memory interface coupled to external memory 150, an interface coupled to the data path logic 218, and an interface coupled to the address path logic 212.
In operation, the port arbitration logic 216 executes a fixed arbitration algorithm to select one of the ports 222 for access to the memory 150. Notably, a plurality of the ports 222 may provide memory transaction requests to the port arbitration logic 216 simultaneously. The port arbitration logic 216 analyzes all pending transaction requests and provides a request acknowledgment to one of the ports 222 in accordance with the fixed arbitration algorithm. The port that “wins” then obtains access to external memory 150 and the requested memory transaction is performed. The port arbitration logic 216 provides port select data to each of the address path logic 212, the data path logic 218, and the control logic 214. The port select data includes the identity of the selected one of the ports 222. The address path logic 212 receives an address context from the selected one of the ports 222 using the port select data. Likewise, the data path logic 218 receives a data context from the selected one of the ports 222 using the port select data.
After granting a transaction request from one of the ports 222, the port arbitration logic 216 provides a memory transaction request to the control logic 214. The control logic 214 processes the memory transaction request and determines a sequence of sub-transactions required to perform the desired memory transaction. Each of the sub-transactions comprises a sequence of memory operations for causing external memory 150 to perform a particular action. Thus, each memory transaction includes a sequence of memory operations.
The control logic 214 drives the data path logic 218, the address path logic 212, and external memory 150 with control signals that execute memory operations in external memory 150. The data path logic 218 drives external memory 150 with data needed to perform the memory operations indicated by the control signals from the control logic 214. Likewise, the address path logic 212 drives external memory 150 with addresses needed to perform the memory operations indicated by the control signals from the control logic 214. The end result is that the requested memory transaction provided by the arbitration logic 216 is performed. The control logic 214 provides a complete signal to the port arbitration logic 216 to indicate when another memory transaction may be issued. In the example of
The ports 340 0-340 N are further connected to Port Interface Modules (PIMs) 330 0-330 N, illustrating that each port is changeable in what it can do, relative to the fixed type ports 222 0-222 3 of
In addition to the ports 340 0-340 N that provide an interface to components external to the MPMC, the memory controller 300 of
Similar to the address path logic 212, control logic 214, arbiter 216 and data path 218 of
The control state machine 314 receives the control information from arbiter 316 and includes circuitry to look up control data needed in BRAM 414. The control signals from BRAM 414 are provided through SRLs 416 to the address path logic 312, data path logic 318, and through a physical (PHY) interface 310 to memory 350. The control signals from the SRLs 416 of the control state machine 316 include port select data that identifies the selected port that will transmit data. The address path logic 312 receives an address context from the selected one of the ports. Likewise, the data path logic 318 receives a data context from the selected port.
After granting a transaction request from one of the input ports, the port arbitration logic arbiter 316 provides a memory transaction request to the control state machine logic 314. The control state machine 314 processes the memory transaction request and determines a sequence of sub-transactions required to perform the desired memory transaction. Each of the sub-transactions comprises a sequence of memory operations for causing the memory to perform a particular action. The control state machine logic 314 drives the data path logic 318, the address path logic 312, and the PHY interface 310 with control signals directed to the memory 350 to execute memory operations. The data path logic 318 and address path logic 312 also drive the PHY interface 310 with data and address signals to perform the memory operations indicated by the control signals from the control state machine logic 314. The end result is that a requested memory transaction provided by the port arbitration logic arbiter 316 is performed.
Self Aligning State Machine
In the control state machine 314, each of ports 0-N has an assigned port address input module 510 0-N. The port 0-N address modules 510 0-N each include a transaction encoder 514 that chooses which sequence will be used in control BRAM 414. The transaction encoder 514 provides a start address based upon the transaction type requested by the port. The start address is provided to register 516 and then to multiplexer 512. Multiplexer 512 selects ‘which port’ is to be granted access from arbiter 316, and then provides the selected start address through optional pipeline register 538 to counter 530. Counter 530 then loads the start address when arbiter 316 tells it to do so. This then begins the sequence of operations to the memory. More details of the arbiter, particularly on how sequences are stored and selected for individual ports, is described in U.S. patent application Ser. No. 11/341,003 referenced previously.
The BRAM 414 of the control path state machine 314 can have contents set to any number of values to permit any kind of control that a user requires. The state machine BRAM 414 provides information with several output paths that control the data path logic 318, address path logic 312 and PHY interface 310 as shown and described with respect to
The basic function of the control state machine 314 is to play sequences of events. The BRAM control path in control state machine 314, in one exemplary embodiment can allow up to 16 sequences. The sequences can include: (1) word write, (2) word read, (3) 4-word cache-line write, (4) 4-word cache-line read, (5) 8-word cache-line write, (6) 8-word cache-line read, (7) 32-word burst write, (8) 32-word burst read, (9) 64-word burst write, (10) 64-word burst read, (11) No operation (NOP), (12) Memory refresh, and (13) Memory initialization.
With the memory 350 being DDR2, DDR, SDRAM, and potentially other types that are driven by a memory controller, the read and write sequences can be divided into three stages: activate, read or write and precharge. Each of these stages has specific sequences of events to the data path 318, address path 312; PHY interface 310, and the arbiter 316 of
Counter 530 generates an address for the BRAM 414 and is used to play a sequence of events from BRAM 414. One of the port 0-n ADDR inputs 510 0-n determines the base address of the selected sequence and is loaded into counter 530. The arbiter 316 contains a prioritized list of port numbers that have access to the control path BRAM state machine 414. Arbiter 316 will look at which ports are requesting access to the state machine and will determine when a new sequence will start and will send the ‘Start Sequence Indicator’ to load counter 530. Additionally, arbiter 316 will present the selected port to the address and data path via the ‘Which Port’ signal. The SRLs 416 provide output words that can be individually delayed by static and/or dynamic control in order to control different functions.
After a sequence is begun to access BRAM 414, in every clock cycle the address increments by one unless the sequence calls for a stall (Cntrl_Stall) or until the sequence has finished and the state machine is ready to accept the next sequence. Each sequence continues until (Cntrl_Complete) is received from one of the SRLs 416 indicating the sequence has finished and the state machine is ready to accept the next sequence. The Cntrl_Stall and Control_Complete signals are provided as feedback through logic 541 to provide an increment (Inc) input to counter 530 to control and hold the state of counter 530 for a specified time until a sequence is complete. Once a sequence is indicated to be complete, the start sequence indicator ‘load’ signal can be provided from the arbiter 316 to indicate that a new sequence base address can be loaded and the next sequence can begin. Although Cntrl_Stall and Cntrl_Complete are illustrated and described, they show one embodiment for implementation of the system and are not specifically required. For example, the sequences could have a fixed length and timing or the high address in the sequence could be used to determine when an operation is complete. It can be appreciated by one of ordinary skill in the art that other types of control signals may be possible and that the above description is provided with the intent to illustrate the concepts.
In one embodiment, the contents of the BRAM 414 can be read or written over a Device Control Register (DCR) bus. A processor included in the MPMC system with a DCR interface can, thus, be used to read and write the contents of the BRAM 414. As illustrated, processor signals are provided over the DCR bus to a B-side interface of BRAM 414. The DCR effectively provides an auto-incrementing keyhole register access to the BRAM 414, as described later. Note as well that the same DCR interface shown in
In operation as shown in
In one embodiment, the dynamic ability of the SRLs 416 is used to allow for altering the order of the content coming from control BRAM 414. Through the use of multiple SRLs 416 (not shown) per control BRAM 414 bit, and with the application of dynamic multiplexing logic (not shown), data bits are allowed to be effectively altered in the output. Bits are altered rather than simply delaying bits before data is sent to the PHY interface. For example, bit 544 can be swapped for bit 546, as illustrated in
Each SRL 416 can be individually and dynamically adjusted as needed to provide appropriate delays. The control over the delay may be static and remain unchanged while the SRL 416 is in operation. It may also be dynamic in that other logic within the MPMC may alter the number of cycles that each individual SRL 416 delays in order to accomplish a specific purpose. Those skilled in the art will understand the value of being able to dynamically delay the outputs of the BRAM 414.
Port Interface Modules (PIMs)
PIM Protocol Bridges
The bus bridge 637 is used, when requested, to provide the bus master from the slave bus interface 632 to generate master access via another PIM, such as the exemplary Master/Slave PIM 650. The provision of bus bridge 637 obviates the need for the significantly more complex bridge 750 and 760 as shown in
The master/slave PIM 650 is another example class of PIM. In this case, the master bus interface 655 is directly connected from Bus Bridge 637 through master/slave PIM 650 to master bus interface 655. It should be noted that master/slave PIM 650 could have additional logic contained within it between bus bridge 637 and master bus interface 655 as required by the application at hand.
One skilled in the art will understand that many different embodiments of PIMs other than those shown in
MPMC Systems with Dedicated Bus Interfaces
Other devices 733, 735 and 737 are connected to PLB 731 to illustrate how various master/slave, slave only, or master only devices can be supported, respectively. Device 733 is an exemplary master/slave device. An example of device 733 could be a Gigabit Ethernet controller that contains a DMA engine to move data to and from memory 350 via memory controller 710. Device 735 is an exemplary slave only device. That is, it only responds to bus transactions from masters on PLB 731. An example of device 735 could be a high speed USB serial controller. Device 737 is an exemplary master only device. That is, it only initiates bus transactions to slaves such as memory controller 710 on PLB 731. An example of device 737 could be a Video Controller which only gathers data from memory 350 and displays it on a CRT or LCD screen.
In addition to the processors, devices, and memory controller 710, PLB 731 illustrates two example bridges that are conventional mechanisms to interconnect PLB 731 and OPB 741. The first, bridge out 750 allows the processors and other masters on PLB 731 to initiate transactions onto OPB 741 where slaves on OPB 741 can respond. The second, bridge in 760 allows master devices on OPB 741 to initiate transactions onto PLB 731, for example transactions to memory 350 via memory controller 710. These bridges are integral to this example system topology because they are the sole means for the masters on the buses to communicate with slaves on other buses. Devices 743, 745 and 747 are analogous to devices 733, 735 and 737, respectively, but are connected to a different protocol, in this case OPB 741.
PIM Embodiments of Invention
A number of advantages can be appreciated from the
Second, by separating the individual entity connections to the MPMC into point-to-point connections, instead of a bus connection, clock frequency of the interface can be improved. In one embodiment, the clock performance of the processor is doubled in speed, improving from 100 MHz to 200 MHz.
Third, by separating the individual entities into point-to-point connections, the devices no longer interfere with one another. That is, in past systems, when devices shared the bus, they had to wait (latency) for access to the bus until other devices completed their access.
Fourth, embodiments of the present invention permit shared bus systems to still be created. The OPB 741 used in
As a fifth advantage, the example embodiment in
As a sixth advantage, similar to elimination of bridge out 750, embodiments of the present invention enable elimination of the bridge in 760 shown in
As a seventh advantage, with embodiments of the present invention eliminating buses and their associated arbiter, arbitration will typically be done inside the memory controller and not within each possible bus. Memory controller performance can then improve because transactions can be more efficiently overlapped in the way that provides the highest possible data rate from the memory 350 using a single arbiter. In shared bus systems, the devices arbitrate for the memory on the bus, and thus the memory cannot take advantage of the parallel knowledge of what transactions are next.
As an eighth advantage, the ports according to embodiments of the present invention can operate at differing frequencies, which permits the ports to match the best operating frequencies of the devices attached. In a typical share bus system as illustrated in
Ninth, the arbitration in arbiter 316 in MPMC 810 is dynamically programmable, so the system performance can be modified as appropriate without the adverse effect of separate bus arbitration. For example, when the system is mainly using the processor(s) to execute code, a first arbitration algorithm can be used, whereas when devices are mainly communicating with memory 350, a second arbitration algorithm can be used that is more efficient for memory communication.
In sum, the structure and location of the bridges in a PIM allow for a wealth of different system topologies. They create high performance systems due to the way they allow transactions to be offloaded from various devices that wish to communicate with memory.
PIMs forming NPI and CDMAC
As indicated previously, with an NPI PIM, the PIM does not need internal logic to translate from the device or processor connected to the port to be compatible with the memory. The Native Port Interface (NPI) devices 902, 904, 906, and 908 can be any kind of device which requires access to memory and which utilizes the NPI protocol to communicate with memory. An example NPI device could be a Video CODEC that captures video from a video input device and sends the video data to memory 350 as well as outputs different video data from memory and sends the video data to a video output device. Since PIMs are programmable and not dedicated to a particular port type, more ports on an MPMC will be available for the typical NPI device as opposed to conventional systems without such dedicated non-NPI interfaces.
An advantage of having DMA engines embedded within the PIMs according to some embodiments of the present invention is that the DMA engine alleviates either the processor or other devices from needing complicated internal DMA engines. The DMA engines can additionally offload the processor connected to its port to another port by directly doing so in a so-called “memcopy” function where memory is copied from one set of memory locations to another set of memory locations without requiring processor access to the memory to perform the copying. This saves significant processing cycles while also having substantially faster execution time to do the copy. In some embodiments, the memcopy function can be done within the PIM using much larger memory transactions than the processor can generate, which further reduces the amount of time the memory is ‘busy’. By minimizing the time the memory is being used by the memcopy, even greater bandwidth is made available to other PIMs that want access to memory 350.
PIM Performance Monitor (PM)
PM Provided Inside PIM
In operation, the PM allows a system to view the performance of a port over time. The PM monitors each transaction and keeps a histogram of the execution time of each type of transaction, including separation for read and write access. The PM can perform a variety of measurements.
The Performance Monitor 1000 also has a per-port performance monitor interface 1001. The performance monitor interface 1001 is generally connected to a processor through a bus interface that is appropriate to that processor. However, it can also be connected to hardware that reads the performance monitor periodically to provide feedback control or other control to the system. In either case, performance monitor interface 1001 is intended as the means of reading and writing the relevant data captured by the performance monitor 1000.
In one embodiment of the present invention, the PM 1000 captures the transactions on the NPI 674 side of the PIM 330. For example, the PM 1000 can capture each byte read separately from each 16-bit word read. The PM 1000 counts how many clock cycles the transaction took to execute, and then accumulates the number of times this type of transaction has occurred at the measured number of clock cycles, providing a histogram of all the types of transactions of various execution times. It should be appreciated by one skilled in the art that many differing types of measurements are possible. For example, the PM 1000 could be built to measure the aggregate data rate on each side of the interfaces.
Since the PM 1000 contains a readable and writable performance monitor interface 1001, and the PM 1000 can be implemented on a per-port basis within the MPMC, the PM 1000 can also act as a control mechanism for dynamically settable parameters with a given PIM 330. That is, if a particular PIM requires some control functions such as setting up a dynamically adjustable base address, the PM 1000 provides a simple means to read or write registers within a given PIM 330 or anywhere else within the MPMC structure.
With a PIM having dynamically programmable arbitration and including per-port PMs, an advantageous dynamic selection of arbitration schemes can be performed. Either software or hardware within the arbiter 316 and control state machine 314 can automatically adjust the arbitration scheme based upon detected performance from a PM to maximize the system level performance.
PM Structure Provided Inside or Outside PIM
In some embodiments of the present invention, the MPMC can have PMs attached to various locations within the MPMC, yet outside the PIM, in order to measure some aspect of performance. For example, in many instances a user wishes to know how much data over how much time has been transferred to/from the memory 350. In one embodiment, the PM would measure the total number of bytes transferred to/from the memory 350 over a specific amount of time. In another embodiment, the PMs are used to measure the length of time that each type of transaction takes to execute. This information is accumulated in a memory within the PM and read out at a later time. By accumulating each type of transaction and the time each transaction time takes into separate ‘buckets’, the PM can contain the data of a histogram of the time each type of transaction takes. This information can be read from the PM's memory either via hardware that affects the state of the MPMC, or via hardware that communicates the information outside the MPMC for a computer or person to look at. It can be appreciated by those skilled in the art that many different types of measurements are possible, including directly measuring the memory 350. The PM then should be understood to not be limited exclusively to the domain shown in
In one embodiment, the BRAM 1006 of the PM has a memory interface bus 1005 which can be coupled to DCR to BRAM connection logic 1004. It should be noted that in some embodiments, memory interface bus 1005 can be attached to one or more BRAMs, such as BRAM 1036 and/or BRAM 1046, which may or may not be contained within a PM or other part of the system. For example, memory interface bus 1005 could be attached to Control BRAM 414 as shown in
The PM's main purpose is to measure performance. In one embodiment, the performance measured is a histogram of the amount of time each type of transaction takes to execute.
The PM of the example embodiment, illustrated in
In some embodiments, the PM uses a DCR bus interface, as shown in
It can be appreciated that the example PM embodiment illustrated in
Configurable Data Path, Address Path, PHY Interface, and Pipelining Data Path
The data path 318, as illustrated in
The size of FIFOs 422, 424 and optional pipeline registers 426, 428 can be adjusted according to embodiments of the invention. To provide efficient operation, the FIFOs for each data port may be changed or adjusted depending on the device or bus attached to a port. Examples of how data path can be morphed or changed are as follows. First, a DSPLB can only do certain transactions, but the DSPLB transactions will dictate minimum FIFO size. Using an SRL based FIFO can be an appropriate choice. In contrast, a PLB can do many transactions and will require a larger storage area, so it likely will be desirable to use BRAM to make the larger FIFO. Lastly, an OPB will typically only use a single word read/write, so a register-type FIFO may be desirable.
For latency control, if FIFO front and back end communicate at different speeds, the type and size of the FIFO can be adjusted in some embodiments of the invention. Previous structures forced fixed type with both end speeds fixed. With adjustable FIFOs, the front and back end speed can be selected depending on attached devices.
FIFOs can further be constructed to accommodate different width memories and NPI sizes according to embodiments of the invention. The optional pipeline registers 426, 428 can be programmably connected from the data path 318 through the PHY interface 310 (which also has registers configurable to accommodate different width memories) to the memory 350 to create a highly configurable data path. Additionally, optional pipeline registers 426, 428 can be used to adjust latencies within the system, act as temporary data storage elements for data realignment when PIMs 330 and memory 350 are of differing size, or act as simple retiming elements to advantageously improve the data path timing. The variable data path allows for different size memory data widths while maintaining a constant interface to the ports and/or different size port widths while maintaining a constant memory data width. Additionally, the configurable data path allows for management of differing clock ratios between the memory 350 and the PIMs 330. For example, changing the memory to PIM clock ratio from 1:1 to 1:2 requires the data path to be in a different physical configuration to properly accumulate and forward data on both sides of the data path. The configurable data path therefore yields high flexibility in implementation area, frequency of operation and architectural functionality.
The programmable data path can also include optional timing management logic (TML) 429 as shown in
In some embodiments, it is preferable to have a single large centralized storage element (e.g., FIFO) per direction between PHY 310 and FIFOs 422, 424. Using optional large centralized FIFOs 425, 427, typically made from a BRAM, allows the FIFO 422 and/or FIFO 424 storage requirements to be lowered. For example, when a large number of ports are used, FIFOs 422 and 424 can consume a significant amount of logic real estate of the overall MPMC. Adding a ‘front-end’ FIFO may allow the FIFOs 422, 424 to be simple register or smaller SRL based FIFOs.
In some embodiments, optional large centralized FIFOs 425 and 427 can serve another advantageous purpose. In many systems, it is desirable to decouple the frequency of memory 350 from the frequency of ports 340. It is further desirable to run memory 350 at the highest possible clock rate that memory 350 is allowed to operate in order to gain additional data bandwidth. Using optional large centralized FIFOs 425, 427 permits the decoupling of frequencies across a single domain. The very high clock rate of memory 350 can have its data path very lightly loaded by directly connecting only to FIFOs 425 and 427. This smaller loading increases the frequency that the memory side of the FIFO may be able to run at.
In yet another embodiment using the optional large centralized FIFOs 425, 427, the data path FIFOs 422, 424 may be able to be removed entirely. By using a BRAM for FIFOs 425, 427, the data from memory 350 can be placed in differing locations within the BRAM corresponding to the port that the data corresponds to. With appropriate adjustments to arbiter 316 and control state machine 314, it is possible to eliminate one or more of the FIFO 422 and/or FIFO 424 from the data path. In such embodiments, the ports ‘share’ the data from the dual ported BRAM. One side attaches to memory 350 via PHY 310 and optional pipeline 426. The other side is shared by the NPI. In another embodiment, multiple BRAMs could be used in order to increase the aggregate data bandwidth possible so that the port side of the FIFOs 425, 427 can be increased. Note that this differs from the normal data path 318 in that there is no longer one FIFO per port per direction. Instead the ports have ‘logical’ FIFOs by their address context within optional centralized FIFOs 425, 427.
Like the data path 318, the address path 312 is also programmable according to embodiments of the invention. The address can be programmed to be multiplexed or non-multiplexed by including multiplexers 415. An SRAM type memory may not use a multiplexed address, while a DRAM type memory may. Thus, the multiplexers 415 of
In some embodiments, per port FIFOs 412 are included in order to allow for transactions to be queued. This gives the MPMC arbiter 316 a priori information to begin arbiting for access to the memory long before it is needed. The result is a substantially more efficient use of both memory 350, and the bus(es) connected to ports 340. In other embodiments, per port FIFOs 412 may be single registers (e.g, a 1 deep FIFO). In still other embodiments, typically where the entire system is fully synchronous at a single rate, FIFOs 412 may be eliminated entirely in order to reduce latency.
In some embodiments, the address size can be dynamically adjusted based on the memory device. In one embodiment, the address size is dynamically changed based on type of memory that is addressed. In this case, additional multiplexers (not shown) are used to ‘reconfigure’ the address to match the memory device requirement. This is a particularly important when two disparate memories are present as part of memory 350. For example, if two DIMMs that have differing addressing requirements are present within the system, accommodation must be made within address path 312 in order to be able to properly accommodate each DIMM.
In some embodiments, the address of each port is independently settable. With adjustable addresses, each port can have address space of its own. However, an overlap between two addresses can also be generated. In some occasions, this is a desirable effect. The address path 312 permits address overlapping, address offset, address replication and address aliasing between ports.
Another configuration demonstrated in
Yet another configuration is shown via replication 1176 in
According to some embodiments of the present invention, provisions are additionally made to prevent collisions between overlapping addresses depending on programming of address size. For example, semaphores may be used to avoid or prevent collisions. This can be true of either an overlap in the physical memory space, and/or each port's logical address space.
The PHY interface 310 according to some embodiments of the present invention is provided that can be parameterized so that data is aligned to match the type of memory connected to the MPMC. The PHY Interface 310 according to some embodiments of the invention is more than a conventional group of registers with clocking. The PHY interface 310 provides for versatile data alignment. Instead of using a state machine to read data and look for a data address strobe (DAS) or column address strobe (CAS) to align data before sending/receiving to/from the PHY interface, embodiments of the PHY layer of the present invention look at data itself.
As shown in
The data alignment state machine 1210 is typically only used during initialization of memory 350. During the training period, state machine 1210 must communicate with multiplexer 1250 to force write data to memory 350 at the appropriate time. The state machine 1210 will control the delay element to properly set the delay upon completion of the training. In one embodiment state machine 1210 begins with the delay set to some value, and steps through each delay value possible until it identifies when data first starts being aligned correctly, and then continues on until it has exhausted the maximum value of the delay, memorizing the points where the data first started to be correct, and where it first stopped being correct. State machine 1210 will then reset the delay 1230 to the midway point between where data first started being correct and where it last was correct. With the PHY embodiments of the present invention, latency is reduced and data alignment is more efficient.
Unlike conventional MPMCs that aligned the data 4 or 8 bits at a time using data strobes, the present PHY embodiment can also align individual bits. With sufficient FPGA input/output hardware, it is also capable of adjusting read and/or write timing on a per-port basis. This is strongly advantageous because it can correct for common mistakes in printed circuit board routing between an FPGA and connected memory device.
Further still, the present PHY interface embodiments are capable of easily connecting to different memory types. While the primary function of the PHY interface is data alignment, it can be appreciated by those skilled in the art that the PHY interface can easily be altered to accommodate a number of different types of memory including SDRAM, DDR SDRAM, DDR2 SDRAM, SRAM, BRAM, RLDRAM, and nearly any other memory technology which the input/outputs of the FPGA can communicate with.
In one embodiment, optional pipeline registers 538 as illustrated in
The optional pipelining of registers shown can likewise be included in other areas of the memory controller. For example, as shown in
In some embodiments of an MPMC of the present invention, the optional pipelines can be dynamically employed as needed. For example, if two disparate DIMMs make up memory 350, one configuration of the optional pipelines can be used when communicating with the first DIMM, whereas a second differing configuration can be used when communicating with a second DIMM. This dynamic ability advantageously provides the best performance for both DIMMs.
In some embodiments, the characteristics of the MPMC for configuring the intelligent pipelining, address path, data path, control path and PHY path are obtained by reading so-called Serial Presence Detect (SPD) Read Only Memory (ROM) typically found on Dual Inline Memory Modules (DIMMs). The SPD ROM is well known in the art. In some embodiments some or all of the variables that affect the MPMC can be read from the SPD ROM, and then enacted in the MPMC structure. This could include configuring the address path, data path, control path, number, type and size of FIFOs used, as well as other memory unique information. In some embodiments, the MPMC's function can be altered dynamically depending upon which of several DIMMs the MPMC is communicating with. This is particularly useful when multiple disparate DIMMs are placed within a system.
Versatile Graphical User Interface
The MPMC implemented using an FPGA can be programmed in some embodiments through a Graphical User Interface (GUI) that provides advantages over prior art. First, the GUI allows system topology creation as well as programming of the MPMC to form a complete project that can be run through Electronic Design Automation (EDA) tools. Previously, GUIs only allowed programming of a single core at one time. With embodiments of the present invention, the GUI can be data driven from user editable text files. It can also offer feedback on resource utilization. Further, it can provide performance feedback estimates based upon the current programmable settings for each port. The GUI can allow users to create sets of data for their configurations of an MPMC and to reuse those for future projects. The GUI uses the information entered to dynamically create an entire core for the MPMC as well as peripherals and processors while providing intelligent design rule checking both as information is being entered, as well as during operation when the GUI dynamically creates complex hardware. The GUI can also be used to display information content from the Performance Monitor(s) within an MPMC system. In some embodiments, this information can be used to dynamically update the arbitration scheme currently in use in the MPMC system. Thus the GUI need not only be used for configuration of the MPMC based system.
In some embodiments of the present invention, different frequencies are allowed to be used within the memory control portion of MPMC. This affects the address path 312, control path state machine 314, port arbiter 316 and in some instances, portions of data path 318. Memory 350 is often a double data rate memory where the data appears on both edges of the clock. There are two typical methods to handle this situation, namely double the output width or double the frequency of the data path. Doubling the frequency of the data path is typically difficult, and thus the usual alternative choice of doubling the output width is selected. However, there is an additional opportunity available to slow down paths within the MPMC based upon some memory technologies. In some instances, including DDR memory, it is possible to run the address and control signals to the memory at ½ the speed of the clock. In some embodiments of the present invention, the PHY interface 310 can contain I/O registers that are clocked at the normal clock rate of the memory, but are fed by information clocked at ½ that rate, and the I/O registers have asynchronous set/reset pins. The control logic (address path 312, control state machine 314, port arbiter 316, and parts of data path 316) is altered as needed to operate at this ½ clock rate of memory 350.
Additionally, some extra logic can be placed within control state machine 314 which runs at the clock frequency of memory 350 and is used to drive the previously mentioned set/reset signals going to the I/O registers within the PHY interface 310. These signals are used to make the signals coming from PHY interface 310 appear as though they are clocked by the memory clock. For example, in DDR memory, there is an activate command followed by a no operation command. The PHY interface would be handed just the activation command at ½ the clock speed of memory 350, but also would be handed the set/reset signals to effect the no operation command at the right time. The net effect of these combinations is the ability to run the memory 350 at much higher effective clock rates than previously possible.
It can be appreciated by one skilled in the art that the logic structures of embodiments described herein could be implemented advantageously using Programmable Logic Device (PLD) technology, more specifically, FPGA style PLDs. However, other implementations are possible using other technologies, such as an Application Specific Integrated Circuit (ASIC), standard cell, or even full custom. These implementations can be “fixed” from the originally programmable implementations in order to accomplish a specific purpose. As such, the present invention need not be limited exclusively to FPGA technology, as one skilled in the art can appreciate.
Although the present invention has been described above with particularity, this was merely to teach one of ordinary skill in the art how to make and use the invention. Many additional modifications will fall within the scope of the invention, as that scope is defined by the following claims.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4349870 *||Sep 5, 1979||Sep 14, 1982||Motorola, Inc.||Microcomputer with programmable multi-function port|
|US4511964 *||Nov 12, 1982||Apr 16, 1985||Hewlett-Packard Company||Dynamic physical memory mapping and management of independent programming environments|
|US4912632||Mar 31, 1988||Mar 27, 1990||International Business Machines Corp.||Memory control subsystem|
|US5651112||Feb 22, 1996||Jul 22, 1997||Hitachi, Ltd.||Information processing system having performance measurement capabilities|
|US5666521 *||Nov 20, 1996||Sep 9, 1997||Intel Corporation||Method and apparatus for performing bit block transfers in a computer system|
|US5761478 *||May 24, 1996||Jun 2, 1998||Texas Instruments Incorporated||Programmable memory interface for efficient transfer of different size data|
|US5768500||Nov 14, 1996||Jun 16, 1998||Lucent Technologies Inc.||Interrupt-based hardware support for profiling memory system performance|
|US5804986 *||Dec 29, 1995||Sep 8, 1998||Cypress Semiconductor Corp.||Memory in a programmable logic device|
|US5870349 *||Oct 28, 1997||Feb 9, 1999||International Business Machines Corporation||Data processing system and method for generating memory control signals with clock skew tolerance|
|US5889987 *||Sep 30, 1996||Mar 30, 1999||Intel Corporation||Flexible non-volatile memory controller with boot block emulation|
|US6151682 *||Sep 4, 1998||Nov 21, 2000||Sarnoff Corporation||Digital signal processing circuitry having integrated timing information|
|US6329839 *||Feb 4, 1999||Dec 11, 2001||Advantage Logic, Inc.||Method and apparatus for universal program controlled bus architecture|
|US6333935||Apr 9, 1998||Dec 25, 2001||Pmc-Sierra Ltd.||State machine architecture with multiplexed random access memory|
|US6353332 *||Feb 7, 2000||Mar 5, 2002||Xilinx, Inc.||Methods for implementing CAM functions using dual-port RAM|
|US6373779 *||May 19, 2000||Apr 16, 2002||Xilinx, Inc.||Block RAM having multiple configurable write modes for use in a field programmable gate array|
|US6567564||Oct 4, 1999||May 20, 2003||Sarnoff Corporation||Pipelined pyramid processor for image processing systems|
|US6601130 *||Nov 24, 1998||Jul 29, 2003||Koninklijke Philips Electronics N.V.||Memory interface unit with programmable strobes to select different memory devices|
|US6601149||Dec 14, 1999||Jul 29, 2003||International Business Machines Corporation||Memory transaction monitoring system and user interface|
|US6662285 *||Jul 27, 2001||Dec 9, 2003||Xilinx, Inc.||User configurable memory system having local and global memory blocks|
|US6754786||Apr 1, 2002||Jun 22, 2004||Mitsubishi Denki Kabushiki Kaisha||Memory control circuit and method for arbitrating memory bus|
|US6792458||Oct 4, 2000||Sep 14, 2004||Urchin Software Corporation||System and method for monitoring and analyzing internet traffic|
|US6799304 *||Oct 1, 2002||Sep 28, 2004||Lsi Logic Corporation||Arbitration within a multiport AMBA slave|
|US6871312||Aug 27, 2002||Mar 22, 2005||Spirent Communications||Method and apparatus for time stamping data|
|US6983393||Dec 11, 2002||Jan 3, 2006||National Instruments Corporation||Deterministically handling asynchronous events in a time triggered system|
|US7020802||Oct 17, 2002||Mar 28, 2006||Sun Microsystems, Inc.||Method and apparatus for monitoring and recording computer system performance parameters|
|US7072800||Sep 19, 2003||Jul 4, 2006||Computer Associates Think, Inc.||Application response monitor|
|US7076595||May 18, 2001||Jul 11, 2006||Xilinx, Inc.||Programmable logic device including programmable interface core and central processing unit|
|US7120761||Oct 31, 2002||Oct 10, 2006||Fujitsu Limited||Multi-port memory based on DRAM core|
|US7146448 *||Sep 21, 2004||Dec 5, 2006||Dot Hill Systems Corporation||Apparatus and method for adopting an orphan I/O port in a redundant storage controller|
|US7225301 *||Nov 20, 2003||May 29, 2007||Quicksilver Technologies||External memory controller node|
|US7260688||Apr 15, 2004||Aug 21, 2007||Xilinx, Inc.||Method and apparatus for controlling access to memory circuitry|
|US7287196||Sep 2, 2004||Oct 23, 2007||International Business Machines Corporation||Measuring reliability of transactions|
|US7308540||May 22, 2003||Dec 11, 2007||Koninklijke Philips Electronics N.V.||Pseudo multiport data memory has stall facility|
|US7308564||Mar 27, 2003||Dec 11, 2007||Xilinx, Inc.||Methods and circuits for realizing a performance monitor for a processor from programmable logic|
|US7454546 *||Jan 27, 2006||Nov 18, 2008||Xilinx, Inc.||Architecture for dynamically reprogrammable arbitration using memory|
|US20030036875||Mar 19, 2002||Feb 20, 2003||Peck Joseph E.||Network-based system for configuring a programmable hardware element in a measurement system using hardware configuration programs generated based on a user specification|
|US20040075690||Dec 12, 2002||Apr 22, 2004||Cirne Lewis K.||User interface for viewing performance information about transactions|
|US20050060456||Sep 16, 2003||Mar 17, 2005||Denali Software, Inc.||Method and apparatus for multi-port memory controller|
|US20060047992||Sep 2, 2004||Mar 2, 2006||International Business Machines Corporation||Measuring reliability of transactions|
|1||*||ARM Limited. ARM PrimeCell MultiPort Memory Controllerl (PL172). Revision r2p1. Technical Reference Manual. 2002.|
|2||Baxter Glenn A. et al. Method and Apparatus for Controlling Access to Memory Circuitry, Serial No. 10/824,967, Filed Apr. 15, 2004, pp. 1-76, available from Xilinx, Inc., 2100 Logic Drive, San Jose, CA 95124.|
|3||Baxter, Glenn A. et al., "Method and Apparatus for Controlling Direct Access to Memory Circuitry", U.S. Appl. No. 10/824,713, filed Apr. 15, 2004, pp. 1-75, available from Xilinx, Inc., 2100 Logic Drive, San Jose, CA 95124.|
|4||Borrelli, Christopher, J. et al., "Method and Apparatus for Communicating Data Between a Network Transceiver and Memory Circuitry", U.S. Appl. No. 10/824,715, filed Apr. 15, 2004, pp. 1-75, available from Xilinx, Inc., 2100 Logic Drive, Jose, CA 95124.|
|5||Lilley, Jennifer R. et al., "Architecture for Dynamically Reprogrammable Arbitration Using Memory", U.S. Appl. No. 11/341,033, filed Jan. 27, 2006, pp. 1-23, available from Xilinx, Inc., 2100 Logic Drive, San Jose, CA 95124.|
|6||*||Microtronix. Avalon Multi-Port SDRAM Controller. User Manual. Version 2.6. Feb. 2010.|
|7||*||STMicroelectronics. Configuring the SPEAr600 multi-port memory controller for external DDR SDRAM. Application Note. Jan. 2010.|
|8||U.S. Appl. No. 11/706,610, filed Feb. 14, 2007 Xilinx, Inc. 2100 Logic Drive, San Jose, CA 95124.|
|9||U.S. Appl. No. 11/707,108, filed Feb. 14, 2007 Xilinx, Inc. 2100 Logic Drive, San Jose, CA 95124.|
|10||U.S. Appl. No. 11/707,175, filed Feb. 14, 2007 Xilinx, Inc. 2100 Logic Drive, San Jose, CA 95124.|
|11||Xilinx, Inc. "Gigabit System Reference Design" XAPP536, Jun. 3, 2004, pp. 1-51, v1.1, available from Xilinx, Inc., 2100 Logic Drive, San Jose, CA 95124.|
|12||Xilinx, Inc., "High Performance Multi-Port Memory Controller", XAPP535, Dec. 10, 2004, pp. 1-190, v1.1, available from Xilinx, Inc., 2100 Logic Drive, San Jose, CA 95124.|
|13||Xilinx, Inc., "Multi Port Memory Controller-MPMC2", Oct. 20, 2006, 2 pp., v.1.7, available from Xilinx, Inc., 2100 Logic Drive, San Jose, CA 95124.|
|14||Xilinx, Inc., "Multi Port Memory Controller—MPMC2", Oct. 20, 2006, 2 pp., v.1.7, available from Xilinx, Inc., 2100 Logic Drive, San Jose, CA 95124.|
|15||*||Zipcores.com. Multi-ported Memory Controller-Arbiter. Rev. 1.0. 2009.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US8245101||Aug 14, 2012||Sandisk Enterprise Ip Llc||Patrol function used in flash storage controller to detect data errors|
|US8365041||Mar 17, 2010||Jan 29, 2013||Sandisk Enterprise Ip Llc||MLC self-raid flash data protection scheme|
|US8386700||Nov 29, 2011||Feb 26, 2013||Sandisk Enterprise Ip Llc||Flash memory controller garbage collection operations performed independently in multiple flash memory groups|
|US8433831 *||Apr 30, 2013||Robert Bosch Gmbh||Method for programming a safety-oriented programmable logic controller and router function block|
|US8452926 *||May 28, 2013||Texas Instruments Incorporated||Memory sharing arrangement|
|US8467218||Dec 13, 2010||Jun 18, 2013||Altera Corporation||System and apparatus with IC resource interconnect|
|US8473814||Jun 27, 2012||Jun 25, 2013||Sandisk Enterprise Ip Llc||MLC self-RAID flash data protection scheme|
|US8484533||Jun 27, 2012||Jul 9, 2013||Sandisk Enterprise Ip Llc||MLC self-RAID flash data protection scheme|
|US8484534||Jun 27, 2012||Jul 9, 2013||Sandisk Enterprise IP LLC.||MLC self-RAID flash data protection scheme|
|US8533384||Apr 8, 2008||Sep 10, 2013||Sandisk Enterprise Ip Llc||Flash memory controller garbage collection operations performed independently in multiple flash memory groups|
|US8543774 *||Apr 5, 2011||Sep 24, 2013||Ansaldo Sts Usa, Inc.||Programmable logic apparatus employing shared memory, vital processor and non-vital communications processor, and system including the same|
|US8621137||Apr 8, 2008||Dec 31, 2013||Sandisk Enterprise Ip Llc||Metadata rebuild in a flash memory controller following a loss of power|
|US8621138||Apr 8, 2008||Dec 31, 2013||Sandisk Enterprise Ip Llc||Flash storage controller execute loop|
|US8738841||Apr 8, 2008||May 27, 2014||Sandisk Enterprise IP LLC.||Flash memory controller and system including data pipelines incorporating multiple buffers|
|US8751755||Apr 8, 2008||Jun 10, 2014||Sandisk Enterprise Ip Llc||Mass storage controller volatile memory containing metadata related to flash memory storage|
|US8762620 *||Apr 8, 2008||Jun 24, 2014||Sandisk Enterprise Ip Llc||Multiprocessor storage controller|
|US8775717||Apr 8, 2008||Jul 8, 2014||Sandisk Enterprise Ip Llc||Storage controller for flash memory including a crossbar switch connecting a plurality of processors with a plurality of internal memories|
|US8793543||Aug 31, 2012||Jul 29, 2014||Sandisk Enterprise Ip Llc||Adaptive read comparison signal generation for memory systems|
|US8885334 *||Mar 10, 2011||Nov 11, 2014||Xilinx, Inc.||Computing system with network attached processors|
|US8891303||Jun 6, 2014||Nov 18, 2014||Sandisk Technologies Inc.||Method and system for dynamic word line based configuration of a three-dimensional memory device|
|US8909982||Oct 31, 2011||Dec 9, 2014||Sandisk Enterprise Ip Llc||System and method for detecting copyback programming problems|
|US8910020||Oct 31, 2011||Dec 9, 2014||Sandisk Enterprise Ip Llc||Intelligent bit recovery for flash memory|
|US8924815||Nov 16, 2012||Dec 30, 2014||Sandisk Enterprise Ip Llc||Systems, methods and devices for decoding codewords having multiple parity segments|
|US8938658||Aug 31, 2012||Jan 20, 2015||Sandisk Enterprise Ip Llc||Statistical read comparison signal generation for memory systems|
|US8954822||Nov 16, 2012||Feb 10, 2015||Sandisk Enterprise Ip Llc||Data encoder and decoder using memory-specific parity-check matrix|
|US8959282||May 10, 2013||Feb 17, 2015||Sandisk Enterprise Ip Llc||Flash storage controller execute loop|
|US8959283||Dec 20, 2013||Feb 17, 2015||Sandisk Enterprise Ip Llc||Flash storage controller execute loop|
|US8972667||Jun 27, 2012||Mar 3, 2015||International Business Machines Corporation||Exchanging data between memory controllers|
|US9003264||Mar 15, 2013||Apr 7, 2015||Sandisk Enterprise Ip Llc||Systems, methods, and devices for multi-dimensional flash RAID data protection|
|US9009576||Mar 15, 2013||Apr 14, 2015||Sandisk Enterprise Ip Llc||Adaptive LLR based on syndrome weight|
|US9043517||Sep 24, 2013||May 26, 2015||Sandisk Enterprise Ip Llc||Multipass programming in buffers implemented in non-volatile data storage systems|
|US9048876||Nov 16, 2012||Jun 2, 2015||Sandisk Enterprise Ip Llc||Systems, methods and devices for multi-tiered error correction|
|US9058289||Aug 31, 2012||Jun 16, 2015||Sandisk Enterprise Ip Llc||Soft information generation for memory systems|
|US9070481||Jun 6, 2014||Jun 30, 2015||Sandisk Technologies Inc.||Internal current measurement for age measurements|
|US9092350||Sep 11, 2013||Jul 28, 2015||Sandisk Enterprise Ip Llc||Detection and handling of unbalanced errors in interleaved codewords|
|US9092370||Dec 19, 2013||Jul 28, 2015||Sandisk Enterprise Ip Llc||Power failure tolerant cryptographic erase|
|US9093160||Jun 6, 2014||Jul 28, 2015||Sandisk Technologies Inc.||Methods and systems for staggered memory operations|
|US9122636||Dec 19, 2013||Sep 1, 2015||Sandisk Enterprise Ip Llc||Hard power fail architecture|
|US9129665||Dec 19, 2013||Sep 8, 2015||Sandisk Enterprise Ip Llc||Dynamic brownout adjustment in a storage device|
|US9136877||Aug 20, 2013||Sep 15, 2015||Sandisk Enterprise Ip Llc||Syndrome layered decoding for LDPC codes|
|US9152555||Nov 15, 2013||Oct 6, 2015||Sandisk Enterprise IP LLC.||Data management with modular erase in a data storage system|
|US9152556||Dec 11, 2013||Oct 6, 2015||Sandisk Enterprise Ip Llc||Metadata rebuild in a flash memory controller following a loss of power|
|US9158349||Dec 19, 2013||Oct 13, 2015||Sandisk Enterprise Ip Llc||System and method for heat dissipation|
|US9158677 *||May 3, 2013||Oct 13, 2015||Sandisk Enterprise Ip Llc||Flash storage controller execute loop|
|US9159437||Jun 21, 2013||Oct 13, 2015||Sandisk Enterprise IP LLC.||Device and method for resolving an LM flag issue|
|US9170941||Nov 15, 2013||Oct 27, 2015||Sandisk Enterprises IP LLC||Data hardening in a storage system|
|US9214965||Nov 8, 2013||Dec 15, 2015||Sandisk Enterprise Ip Llc||Method and system for improving data integrity in non-volatile storage|
|US9235245||Dec 19, 2013||Jan 12, 2016||Sandisk Enterprise Ip Llc||Startup performance and power isolation|
|US9235509||Sep 17, 2013||Jan 12, 2016||Sandisk Enterprise Ip Llc||Write amplification reduction by delaying read access to data written during garbage collection|
|US9236886||Sep 17, 2013||Jan 12, 2016||Sandisk Enterprise Ip Llc||Universal and reconfigurable QC-LDPC encoder|
|US9239751||Mar 15, 2013||Jan 19, 2016||Sandisk Enterprise Ip Llc||Compressing data from multiple reads for error control management in memory systems|
|US9239783||May 14, 2013||Jan 19, 2016||Sandisk Enterprise Ip Llc||Multiprocessor storage controller|
|US9244763||Sep 17, 2013||Jan 26, 2016||Sandisk Enterprise Ip Llc||System and method for updating a reading threshold voltage based on symbol transition information|
|US9244785||Dec 19, 2013||Jan 26, 2016||Sandisk Enterprise Ip Llc||Simulated power failure and data hardening|
|US9244867||Jun 1, 2011||Jan 26, 2016||Altera Corporation||Memory controller interface with adjustable port widths|
|US9250676||Dec 19, 2013||Feb 2, 2016||Sandisk Enterprise Ip Llc||Power failure architecture and verification|
|US9263156||Dec 19, 2013||Feb 16, 2016||Sandisk Enterprise Ip Llc||System and method for adjusting trip points within a storage device|
|US9280429||Dec 19, 2013||Mar 8, 2016||Sandisk Enterprise Ip Llc||Power fail latching based on monitoring multiple power supply voltages in a storage device|
|US9298608||Dec 20, 2013||Mar 29, 2016||Sandisk Enterprise Ip Llc||Biasing for wear leveling in storage systems|
|US9323637||Dec 19, 2013||Apr 26, 2016||Sandisk Enterprise Ip Llc||Power sequencing and data hardening architecture|
|US9329928||Nov 8, 2013||May 3, 2016||Sandisk Enterprise IP LLC.||Bandwidth optimization in a non-volatile memory system|
|US9348377||Apr 3, 2014||May 24, 2016||Sandisk Enterprise Ip Llc||Thermal isolation techniques|
|US9361221||Sep 17, 2013||Jun 7, 2016||Sandisk Technologies Inc.||Write amplification reduction through reliable writes during garbage collection|
|US9367246||Aug 9, 2013||Jun 14, 2016||Sandisk Technologies Inc.||Performance optimization of data transfer for soft information generation|
|US9384126||Sep 24, 2013||Jul 5, 2016||Sandisk Technologies Inc.||Methods and systems to avoid false negative results in bloom filters implemented in non-volatile data storage systems|
|US9390021||Jul 21, 2014||Jul 12, 2016||Sandisk Technologies Llc||Efficient cache utilization in a tiered data structure|
|US9390814||Aug 7, 2014||Jul 12, 2016||Sandisk Technologies Llc||Fault detection and prediction for data storage elements|
|US20090172258 *||Apr 8, 2008||Jul 2, 2009||Pliant Technology, Inc||Flash memory controller garbage collection operations performed independently in multiple flash memory groups|
|US20090172259 *||Apr 8, 2008||Jul 2, 2009||Pliant Technology, Inc.||Mass storage controller volatile memory containing metadata related to flash memory storage|
|US20090172260 *||Apr 8, 2008||Jul 2, 2009||Pliant Technology, Inc.||Flash memory controller and system including data pipelines incorporating multiple buffers|
|US20090172261 *||Apr 8, 2008||Jul 2, 2009||Pliant Technology, Inc.||Multiprocessor storage controller|
|US20090172262 *||Apr 8, 2008||Jul 2, 2009||Pliant Technology, Inc.||Metadata rebuild in a flash memory controller following a loss of power|
|US20090172263 *||Apr 8, 2008||Jul 2, 2009||Pliant Technology, Inc.||Flash storage controller execute loop|
|US20090172308 *||Apr 8, 2008||Jul 2, 2009||Pliant Technology, Inc.||Storage controller for flash memory including a crossbar switch connecting a plurality of processors with a plurality of internal memories|
|US20090172499 *||Apr 8, 2008||Jul 2, 2009||Pliant Technology, Inc.||Patrol function used in flash storage controller to detect data errors|
|US20100138568 *||Nov 30, 2009||Jun 3, 2010||Jochen Ost||Method for programming a safety-oriented programmable logic controller and router function block|
|US20110055490 *||Jan 5, 2010||Mar 3, 2011||Philippe Gentric||Memory Sharing Arrangement|
|US20120260046 *||Oct 11, 2012||Lemonovich John E||Programmable logic apparatus employing shared memory, vital processor and non-vital communications processor, and system including the same|
|US20130339581 *||May 3, 2013||Dec 19, 2013||Sandisk Enterprise Ip Llc||Flash Storage Controller Execute Loop|
|US20150199468 *||Sep 14, 2012||Jul 16, 2015||Freescale Semiconductor, Inc.||Method and apparatus for selecting data path elements for cloning|
|U.S. Classification||710/305, 711/149, 710/315, 326/38, 710/12, 710/306, 710/314, 710/10|
|Feb 14, 2007||AS||Assignment|
Owner name: XILINX, INC., CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:BAXTER, GLENN A.;REEL/FRAME:018984/0922
Effective date: 20070209
|Sep 22, 2014||FPAY||Fee payment|
Year of fee payment: 4